The Design and Implementation of the FreeBSD Operating System, Second Edition
Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]

FreeBSD/Linux Kernel Cross Reference
sys/netgraph/netflow/netflow.c

Version: -  FREEBSD  -  FREEBSD-13-STABLE  -  FREEBSD-13-0  -  FREEBSD-12-STABLE  -  FREEBSD-12-0  -  FREEBSD-11-STABLE  -  FREEBSD-11-0  -  FREEBSD-10-STABLE  -  FREEBSD-10-0  -  FREEBSD-9-STABLE  -  FREEBSD-9-0  -  FREEBSD-8-STABLE  -  FREEBSD-8-0  -  FREEBSD-7-STABLE  -  FREEBSD-7-0  -  FREEBSD-6-STABLE  -  FREEBSD-6-0  -  FREEBSD-5-STABLE  -  FREEBSD-5-0  -  FREEBSD-4-STABLE  -  FREEBSD-3-STABLE  -  FREEBSD22  -  l41  -  OPENBSD  -  linux-2.6  -  MK84  -  PLAN9  -  xnu-8792 
SearchContext: -  none  -  3  -  10 

    1 /*-
    2  * Copyright (c) 2004-2005 Gleb Smirnoff <glebius@FreeBSD.org>
    3  * Copyright (c) 2001-2003 Roman V. Palagin <romanp@unshadow.net>
    4  * All rights reserved.
    5  *
    6  * Redistribution and use in source and binary forms, with or without
    7  * modification, are permitted provided that the following conditions
    8  * are met:
    9  * 1. Redistributions of source code must retain the above copyright
   10  *    notice, this list of conditions and the following disclaimer.
   11  * 2. Redistributions in binary form must reproduce the above copyright
   12  *    notice, this list of conditions and the following disclaimer in the
   13  *    documentation and/or other materials provided with the distribution.
   14  *
   15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
   16  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
   17  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
   18  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
   19  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
   20  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
   21  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
   22  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
   23  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
   24  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
   25  * SUCH DAMAGE.
   26  *
   27  * $SourceForge: netflow.c,v 1.41 2004/09/05 11:41:10 glebius Exp $
   28  */
   29 
   30 static const char rcs_id[] =
   31     "@(#) $FreeBSD: releng/6.0/sys/netgraph/netflow/netflow.c 146285 2005-05-16 17:10:08Z glebius $";
   32 
   33 #include <sys/param.h>
   34 #include <sys/kernel.h>
   35 #include <sys/limits.h>
   36 #include <sys/mbuf.h>
   37 #include <sys/syslog.h>
   38 #include <sys/systm.h>
   39 #include <sys/socket.h>
   40 
   41 #include <machine/atomic.h>
   42 
   43 #include <net/if.h>
   44 #include <net/route.h>
   45 #include <netinet/in.h>
   46 #include <netinet/in_systm.h>
   47 #include <netinet/ip.h>
   48 #include <netinet/tcp.h>
   49 #include <netinet/udp.h>
   50 
   51 #include <netgraph/ng_message.h>
   52 #include <netgraph/netgraph.h>
   53 
   54 #include <netgraph/netflow/netflow.h>
   55 #include <netgraph/netflow/ng_netflow.h>
   56 
   57 #define NBUCKETS        (65536)         /* must be power of 2 */
   58 
   59 /* This hash is for TCP or UDP packets */
   60 #define FULL_HASH(addr1,addr2,port1,port2)\
   61         (((addr1 >> 16) ^               \
   62           (addr2 & 0x00FF) ^            \
   63           ((port1 ^ port2) << 8) )&     \
   64          (NBUCKETS - 1))
   65 
   66 /* This hash is for all other IP packets */
   67 #define ADDR_HASH(addr1,addr2)\
   68         (((addr1 >> 16) ^               \
   69           (addr2 & 0x00FF) )&           \
   70          (NBUCKETS - 1))
   71 
   72 /* Macros to shorten logical constructions */
   73 /* XXX: priv must exist in namespace */
   74 #define INACTIVE(fle)   (time_uptime - fle->f.last > priv->info.nfinfo_inact_t)
   75 #define AGED(fle)       (time_uptime - fle->f.first > priv->info.nfinfo_act_t)
   76 #define ISFREE(fle)     (fle->f.packets == 0)
   77 
   78 /*
   79  * 4 is a magical number: statistically number of 4-packet flows is
   80  * bigger than 5,6,7...-packet flows by an order of magnitude. Most UDP/ICMP
   81  * scans are 1 packet (~ 90% of flow cache). TCP scans are 2-packet in case
   82  * of reachable host and 4-packet otherwise.
   83  */
   84 #define SMALL(fle)      (fle->f.packets <= 4)
   85 
   86 /*
   87  * Cisco uses milliseconds for uptime. Bad idea, since it overflows
   88  * every 48+ days. But we will do same to keep compatibility. This macro
   89  * does overflowable multiplication to 1000.
   90  */
   91 #define MILLIUPTIME(t)  (((t) << 9) +   /* 512 */       \
   92                          ((t) << 8) +   /* 256 */       \
   93                          ((t) << 7) +   /* 128 */       \
   94                          ((t) << 6) +   /* 64  */       \
   95                          ((t) << 5) +   /* 32  */       \
   96                          ((t) << 3))    /* 8   */
   97 
   98 MALLOC_DECLARE(M_NETFLOW_HASH);
   99 MALLOC_DEFINE(M_NETFLOW_HASH, "NetFlow hash", "NetFlow hash");
  100 
  101 static int export_add(item_p, struct flow_entry *);
  102 static int export_send(priv_p, item_p);
  103 
  104 /* Generate hash for a given flow record. */
  105 static __inline uint32_t
  106 ip_hash(struct flow_rec *r)
  107 {
  108         switch (r->r_ip_p) {
  109         case IPPROTO_TCP:
  110         case IPPROTO_UDP:
  111                 return FULL_HASH(r->r_src.s_addr, r->r_dst.s_addr,
  112                     r->r_sport, r->r_dport);
  113         default:
  114                 return ADDR_HASH(r->r_src.s_addr, r->r_dst.s_addr);
  115         }
  116 }
  117 
  118 /* This is callback from uma(9), called on alloc. */
  119 static int
  120 uma_ctor_flow(void *mem, int size, void *arg, int how)
  121 {
  122         priv_p priv = (priv_p )arg;
  123 
  124         if (atomic_load_acq_32(&priv->info.nfinfo_used) >= CACHESIZE)
  125                 return (ENOMEM);
  126 
  127         atomic_add_32(&priv->info.nfinfo_used, 1);
  128 
  129         return (0);
  130 }
  131 
  132 /* This is callback from uma(9), called on free. */
  133 static void
  134 uma_dtor_flow(void *mem, int size, void *arg)
  135 {
  136         priv_p priv = (priv_p )arg;
  137 
  138         atomic_subtract_32(&priv->info.nfinfo_used, 1);
  139 }
  140 
  141 /*
  142  * Detach export datagram from priv, if there is any.
  143  * If there is no, allocate a new one.
  144  */
  145 static item_p
  146 get_export_dgram(priv_p priv)
  147 {
  148         item_p  item = NULL;
  149 
  150         mtx_lock(&priv->export_mtx);
  151         if (priv->export_item != NULL) {
  152                 item = priv->export_item;
  153                 priv->export_item = NULL;
  154         }
  155         mtx_unlock(&priv->export_mtx);
  156 
  157         if (item == NULL) {
  158                 struct netflow_v5_export_dgram *dgram;
  159                 struct mbuf *m;
  160 
  161                 m = m_getcl(M_DONTWAIT, MT_DATA, M_PKTHDR);
  162                 if (m == NULL)
  163                         return (NULL);
  164                 item = ng_package_data(m, NG_NOFLAGS);
  165                 if (item == NULL)
  166                         return (NULL);
  167                 dgram = mtod(m, struct netflow_v5_export_dgram *);
  168                 dgram->header.count = 0;
  169                 dgram->header.version = htons(NETFLOW_V5);
  170 
  171         }
  172 
  173         return (item);
  174 }
  175 
  176 /*
  177  * Re-attach incomplete datagram back to priv.
  178  * If there is already another one, then send incomplete. */
  179 static void
  180 return_export_dgram(priv_p priv, item_p item)
  181 {
  182         /*
  183          * It may happen on SMP, that some thread has already
  184          * put its item there, in this case we bail out and
  185          * send what we have to collector.
  186          */
  187         mtx_lock(&priv->export_mtx);
  188         if (priv->export_item == NULL) {
  189                 priv->export_item = item;
  190                 mtx_unlock(&priv->export_mtx);
  191         } else {
  192                 mtx_unlock(&priv->export_mtx);
  193                 export_send(priv, item);
  194         }
  195 }
  196 
  197 /*
  198  * The flow is over. Call export_add() and free it. If datagram is
  199  * full, then call export_send().
  200  */
  201 static __inline void
  202 expire_flow(priv_p priv, item_p *item, struct flow_entry *fle)
  203 {
  204         if (*item == NULL)
  205                 *item = get_export_dgram(priv);
  206         if (*item == NULL) {
  207                 atomic_add_32(&priv->info.nfinfo_export_failed, 1);
  208                 uma_zfree_arg(priv->zone, fle, priv);
  209                 return;
  210         }
  211         if (export_add(*item, fle) > 0) {
  212                 export_send(priv, *item);
  213                 *item = NULL;
  214         }
  215         uma_zfree_arg(priv->zone, fle, priv);
  216 }
  217 
  218 /* Get a snapshot of node statistics */
  219 void
  220 ng_netflow_copyinfo(priv_p priv, struct ng_netflow_info *i)
  221 {
  222         /* XXX: atomic */
  223         memcpy((void *)i, (void *)&priv->info, sizeof(priv->info));
  224 }
  225 
  226 /* Calculate number of bits in netmask */
  227 #define g21     0x55555555ul    /* = 0101_0101_0101_0101_0101_0101_0101_0101 */
  228 #define g22     0x33333333ul    /* = 0011_0011_0011_0011_0011_0011_0011_0011 */
  229 #define g23     0x0f0f0f0ful    /* = 0000_1111_0000_1111_0000_1111_0000_1111 */
  230 static __inline u_char
  231 bit_count(uint32_t v)
  232 {
  233         v = (v & g21) + ((v >> 1) & g21);
  234         v = (v & g22) + ((v >> 2) & g22);
  235         v = (v + (v >> 4)) & g23;
  236         return (v + (v >> 8) + (v >> 16) + (v >> 24)) & 0x3f;
  237 }
  238 
  239 /*
  240  * Insert a record into defined slot.
  241  *
  242  * First we get for us a free flow entry, then fill in all
  243  * possible fields in it.
  244  *
  245  * TODO: consider dropping hash mutex while filling in datagram,
  246  * as this was done in previous version. Need to test & profile
  247  * to be sure.
  248  */
  249 static __inline int
  250 hash_insert(priv_p priv, struct flow_hash_entry  *hsh, struct flow_rec *r,
  251         int plen, uint8_t tcp_flags)
  252 {
  253         struct flow_entry       *fle;
  254         struct route ro;
  255         struct sockaddr_in *sin;
  256 
  257         mtx_assert(&hsh->mtx, MA_OWNED);
  258 
  259         fle = uma_zalloc_arg(priv->zone, priv, M_NOWAIT);
  260         if (fle == NULL) {
  261                 atomic_add_32(&priv->info.nfinfo_alloc_failed, 1);
  262                 return (ENOMEM);
  263         }
  264 
  265         /*
  266          * Now fle is totally ours. It is detached from all lists,
  267          * we can safely edit it.
  268          */
  269 
  270         bcopy(r, &fle->f.r, sizeof(struct flow_rec));
  271         fle->f.bytes = plen;
  272         fle->f.packets = 1;
  273         fle->f.tcp_flags = tcp_flags;
  274 
  275         fle->f.first = fle->f.last = time_uptime;
  276 
  277         /*
  278          * First we do route table lookup on destination address. So we can
  279          * fill in out_ifx, dst_mask, nexthop, and dst_as in future releases.
  280          */
  281         bzero((caddr_t)&ro, sizeof(ro));
  282         sin = (struct sockaddr_in *)&ro.ro_dst;
  283         sin->sin_len = sizeof(*sin);
  284         sin->sin_family = AF_INET;
  285         sin->sin_addr = fle->f.r.r_dst;
  286         rtalloc_ign(&ro, RTF_CLONING);
  287         if (ro.ro_rt != NULL) {
  288                 struct rtentry *rt = ro.ro_rt;
  289 
  290                 fle->f.fle_o_ifx = rt->rt_ifp->if_index;
  291 
  292                 if (rt->rt_flags & RTF_GATEWAY &&
  293                     rt->rt_gateway->sa_family == AF_INET)
  294                         fle->f.next_hop =
  295                             ((struct sockaddr_in *)(rt->rt_gateway))->sin_addr;
  296 
  297                 if (rt_mask(rt))
  298                         fle->f.dst_mask =
  299                             bit_count(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
  300                 else if (rt->rt_flags & RTF_HOST)
  301                         /* Give up. We can't determine mask :( */
  302                         fle->f.dst_mask = 32;
  303 
  304                 RTFREE(ro.ro_rt);
  305         }
  306 
  307         /* Do route lookup on source address, to fill in src_mask. */
  308 
  309         bzero((caddr_t)&ro, sizeof(ro));
  310         sin = (struct sockaddr_in *)&ro.ro_dst;
  311         sin->sin_len = sizeof(*sin);
  312         sin->sin_family = AF_INET;
  313         sin->sin_addr = fle->f.r.r_src;
  314         rtalloc_ign(&ro, RTF_CLONING);
  315         if (ro.ro_rt != NULL) {
  316                 struct rtentry *rt = ro.ro_rt;
  317 
  318                 if (rt_mask(rt))
  319                         fle->f.src_mask =
  320                             bit_count(((struct sockaddr_in *)rt_mask(rt))->sin_addr.s_addr);
  321                 else if (rt->rt_flags & RTF_HOST)
  322                         /* Give up. We can't determine mask :( */
  323                         fle->f.src_mask = 32;
  324 
  325                 RTFREE(ro.ro_rt);
  326         }
  327 
  328         /* Push new flow at the and of hash. */
  329         TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
  330 
  331         return (0);
  332 }
  333 
  334 
  335 /*
  336  * Non-static functions called from ng_netflow.c
  337  */
  338 
  339 /* Allocate memory and set up flow cache */
  340 int
  341 ng_netflow_cache_init(priv_p priv)
  342 {
  343         struct flow_hash_entry  *hsh;
  344         int i;
  345 
  346         /* Initialize cache UMA zone. */
  347         priv->zone = uma_zcreate("NetFlow cache", sizeof(struct flow_entry),
  348             uma_ctor_flow, uma_dtor_flow, NULL, NULL, UMA_ALIGN_CACHE, 0);
  349         uma_zone_set_max(priv->zone, CACHESIZE);
  350 
  351         /* Allocate hash. */
  352         MALLOC(priv->hash, struct flow_hash_entry *,
  353             NBUCKETS * sizeof(struct flow_hash_entry),
  354             M_NETFLOW_HASH, M_WAITOK | M_ZERO);
  355 
  356         if (priv->hash == NULL) {
  357                 uma_zdestroy(priv->zone);
  358                 return (ENOMEM);
  359         }
  360 
  361         /* Initialize hash. */
  362         for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++) {
  363                 mtx_init(&hsh->mtx, "hash mutex", NULL, MTX_DEF);
  364                 TAILQ_INIT(&hsh->head);
  365         }
  366 
  367         mtx_init(&priv->export_mtx, "export dgram lock", NULL, MTX_DEF);
  368 
  369         return (0);
  370 }
  371 
  372 /* Free all flow cache memory. Called from node close method. */
  373 void
  374 ng_netflow_cache_flush(priv_p priv)
  375 {
  376         struct flow_entry       *fle, *fle1;
  377         struct flow_hash_entry  *hsh;
  378         item_p                  item = NULL;
  379         int i;
  380 
  381         /*
  382          * We are going to free probably billable data.
  383          * Expire everything before freeing it.
  384          * No locking is required since callout is already drained.
  385          */
  386         for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++)
  387                 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
  388                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  389                         expire_flow(priv, &item, fle);
  390                 }
  391 
  392         if (item != NULL)
  393                 export_send(priv, item);
  394 
  395         uma_zdestroy(priv->zone);
  396 
  397         /* Destroy hash mutexes. */
  398         for (i = 0, hsh = priv->hash; i < NBUCKETS; i++, hsh++)
  399                 mtx_destroy(&hsh->mtx);
  400 
  401         /* Free hash memory. */
  402         if (priv->hash)
  403                 FREE(priv->hash, M_NETFLOW_HASH);
  404 
  405         mtx_destroy(&priv->export_mtx);
  406 }
  407 
  408 /* Insert packet from into flow cache. */
  409 int
  410 ng_netflow_flow_add(priv_p priv, struct ip *ip, iface_p iface,
  411         struct ifnet *ifp)
  412 {
  413         register struct flow_entry      *fle, *fle1;
  414         struct flow_hash_entry          *hsh;
  415         struct flow_rec         r;
  416         item_p                  item = NULL;
  417         int                     hlen, plen;
  418         int                     error = 0;
  419         uint8_t                 tcp_flags = 0;
  420 
  421         /* Try to fill flow_rec r */
  422         bzero(&r, sizeof(r));
  423         /* check version */
  424         if (ip->ip_v != IPVERSION)
  425                 return (EINVAL);
  426 
  427         /* verify min header length */
  428         hlen = ip->ip_hl << 2;
  429 
  430         if (hlen < sizeof(struct ip))
  431                 return (EINVAL);
  432 
  433         r.r_src = ip->ip_src;
  434         r.r_dst = ip->ip_dst;
  435 
  436         /* save packet length */
  437         plen = ntohs(ip->ip_len);
  438 
  439         r.r_ip_p = ip->ip_p;
  440         r.r_tos = ip->ip_tos;
  441 
  442         /* Configured in_ifx overrides mbuf's */
  443         if (iface->info.ifinfo_index == 0) {
  444                 if (ifp != NULL)
  445                         r.r_i_ifx = ifp->if_index;
  446         } else
  447                 r.r_i_ifx = iface->info.ifinfo_index;
  448 
  449         /*
  450          * XXX NOTE: only first fragment of fragmented TCP, UDP and
  451          * ICMP packet will be recorded with proper s_port and d_port.
  452          * Following fragments will be recorded simply as IP packet with
  453          * ip_proto = ip->ip_p and s_port, d_port set to zero.
  454          * I know, it looks like bug. But I don't want to re-implement
  455          * ip packet assebmling here. Anyway, (in)famous trafd works this way -
  456          * and nobody complains yet :)
  457          */
  458         if ((ip->ip_off & htons(IP_OFFMASK)) == 0)
  459                 switch(r.r_ip_p) {
  460                 case IPPROTO_TCP:
  461                 {
  462                         register struct tcphdr *tcp;
  463 
  464                         tcp = (struct tcphdr *)((caddr_t )ip + hlen);
  465                         r.r_sport = tcp->th_sport;
  466                         r.r_dport = tcp->th_dport;
  467                         tcp_flags = tcp->th_flags;
  468                         break;
  469                 }
  470                         case IPPROTO_UDP:
  471                         r.r_ports = *(uint32_t *)((caddr_t )ip + hlen);
  472                         break;
  473                 }
  474 
  475         /* Update node statistics. XXX: race... */
  476         priv->info.nfinfo_packets ++;
  477         priv->info.nfinfo_bytes += plen;
  478 
  479         /* Find hash slot. */
  480         hsh = &priv->hash[ip_hash(&r)];
  481 
  482         mtx_lock(&hsh->mtx);
  483 
  484         /*
  485          * Go through hash and find our entry. If we encounter an
  486          * entry, that should be expired, purge it. We do a reverse
  487          * search since most active entries are first, and most
  488          * searches are done on most active entries.
  489          */
  490         TAILQ_FOREACH_REVERSE_SAFE(fle, &hsh->head, fhead, fle_hash, fle1) {
  491                 if (bcmp(&r, &fle->f.r, sizeof(struct flow_rec)) == 0)
  492                         break;
  493                 if ((INACTIVE(fle) && SMALL(fle)) || AGED(fle)) {
  494                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  495                         expire_flow(priv, &item, fle);
  496                         atomic_add_32(&priv->info.nfinfo_act_exp, 1);
  497                 }
  498         }
  499 
  500         if (fle) {                      /* An existent entry. */
  501 
  502                 fle->f.bytes += plen;
  503                 fle->f.packets ++;
  504                 fle->f.tcp_flags |= tcp_flags;
  505                 fle->f.last = time_uptime;
  506 
  507                 /*
  508                  * We have the following reasons to expire flow in active way:
  509                  * - it hit active timeout
  510                  * - a TCP connection closed
  511                  * - it is going to overflow counter
  512                  */
  513                 if (tcp_flags & TH_FIN || tcp_flags & TH_RST || AGED(fle) ||
  514                     (fle->f.bytes >= (UINT_MAX - IF_MAXMTU)) ) {
  515                         TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  516                         expire_flow(priv, &item, fle);
  517                 } else {
  518                         /*
  519                          * It is the newest, move it to the tail,
  520                          * if it isn't there already. Next search will
  521                          * locate it quicker.
  522                          */
  523                         if (fle != TAILQ_LAST(&hsh->head, fhead)) {
  524                                 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  525                                 TAILQ_INSERT_TAIL(&hsh->head, fle, fle_hash);
  526                         }
  527                 }
  528         } else                          /* A new flow entry. */
  529                 error = hash_insert(priv, hsh, &r, plen, tcp_flags);
  530 
  531         mtx_unlock(&hsh->mtx);
  532 
  533         if (item != NULL)
  534                 return_export_dgram(priv, item);
  535 
  536         return (error);
  537 }
  538 
  539 /*
  540  * Return records from cache to userland.
  541  *
  542  * TODO: consider NGM_READONLY
  543  * TODO: matching particular IP should be done in kernel, here.
  544  */
  545 int
  546 ng_netflow_flow_show(priv_p priv, uint32_t last, struct ng_mesg *resp)
  547 {
  548         struct flow_hash_entry *hsh;
  549         struct flow_entry *fle;
  550         struct ngnf_flows *data;
  551         int i;
  552 
  553         data = (struct ngnf_flows *)resp->data;
  554         data->last = 0;
  555         data->nentries = 0;
  556 
  557         /* Check if this is a first run */
  558         if (last == 0) {
  559                 hsh = priv->hash;
  560                 i = 0;
  561         } else {
  562                 if (last > NBUCKETS-1)
  563                         return (EINVAL);
  564                 hsh = priv->hash + last;
  565                 i = last;
  566         }
  567 
  568         /*
  569          * We will transfer not more than NREC_AT_ONCE. More data
  570          * will come in next message.
  571          * We send current hash index to userland, and userland should
  572          * return it back to us. Then, we will restart with new entry.
  573          *
  574          * The resulting cache snapshot is inaccurate for the
  575          * following reasons:
  576          *  - we skip locked hash entries
  577          *  - we bail out, if someone wants our entry
  578          *  - we skip rest of entry, when hit NREC_AT_ONCE
  579          */
  580         for (; i < NBUCKETS; hsh++, i++) {
  581                 if (mtx_trylock(&hsh->mtx) == 0)
  582                         continue;
  583 
  584                 TAILQ_FOREACH(fle, &hsh->head, fle_hash) {
  585                         if (hsh->mtx.mtx_lock & MTX_CONTESTED)
  586                                 break;
  587 
  588                         bcopy(&fle->f, &(data->entries[data->nentries]),
  589                             sizeof(fle->f));
  590                         data->nentries++;
  591                         if (data->nentries == NREC_AT_ONCE) {
  592                                 mtx_unlock(&hsh->mtx);
  593                                 if (++i < NBUCKETS)
  594                                         data->last = i;
  595                                 return (0);
  596                         }
  597                 }
  598                 mtx_unlock(&hsh->mtx);
  599         }
  600 
  601         return (0);
  602 }
  603 
  604 /* We have full datagram in privdata. Send it to export hook. */
  605 static int
  606 export_send(priv_p priv, item_p item)
  607 {
  608         struct mbuf *m = NGI_M(item);
  609         struct netflow_v5_export_dgram *dgram = mtod(m,
  610                                         struct netflow_v5_export_dgram *);
  611         struct netflow_v5_header *header = &dgram->header;
  612         struct timespec ts;
  613         int error = 0;
  614 
  615         /* Fill mbuf header. */
  616         m->m_len = m->m_pkthdr.len = sizeof(struct netflow_v5_record) *
  617            header->count + sizeof(struct netflow_v5_header);
  618 
  619         /* Fill export header. */
  620         header->sys_uptime = htonl(MILLIUPTIME(time_uptime));
  621         getnanotime(&ts);
  622         header->unix_secs  = htonl(ts.tv_sec);
  623         header->unix_nsecs = htonl(ts.tv_nsec);
  624         header->count = htons(header->count);
  625         header->flow_seq = htonl(atomic_load_acq_32(&priv->flow_seq));
  626 
  627         /* Flow sequence contains number of first record, so it
  628            is updated after being put in header. */
  629         atomic_add_32(&priv->flow_seq, header->count);
  630 
  631         if (priv->export != NULL)
  632                 /* Should also NET_LOCK_GIANT(). */
  633                 NG_FWD_ITEM_HOOK(error, item, priv->export);
  634 
  635         return (error);
  636 }
  637 
  638 
  639 /* Add export record to dgram. */
  640 static int
  641 export_add(item_p item, struct flow_entry *fle)
  642 {
  643         struct netflow_v5_export_dgram *dgram = mtod(NGI_M(item),
  644                                         struct netflow_v5_export_dgram *);
  645         struct netflow_v5_header *header = &dgram->header;
  646         struct netflow_v5_record *rec;
  647 
  648         if (header->count == 0 ) {      /* first record */
  649                 rec = &dgram->r[0];
  650                 header->count = 1;
  651         } else {                        /* continue filling datagram */
  652                 rec = &dgram->r[header->count];
  653                 header->count ++;
  654         }
  655 
  656         KASSERT(header->count <= NETFLOW_V5_MAX_RECORDS,
  657             ("ng_netflow: export too big"));
  658 
  659         /* Fill in export record. */
  660         rec->src_addr = fle->f.r.r_src.s_addr;
  661         rec->dst_addr = fle->f.r.r_dst.s_addr;
  662         rec->next_hop = fle->f.next_hop.s_addr;
  663         rec->i_ifx    = htons(fle->f.fle_i_ifx);
  664         rec->o_ifx    = htons(fle->f.fle_o_ifx);
  665         rec->packets  = htonl(fle->f.packets);
  666         rec->octets   = htonl(fle->f.bytes);
  667         rec->first    = htonl(MILLIUPTIME(fle->f.first));
  668         rec->last     = htonl(MILLIUPTIME(fle->f.last));
  669         rec->s_port   = fle->f.r.r_sport;
  670         rec->d_port   = fle->f.r.r_dport;
  671         rec->flags    = fle->f.tcp_flags;
  672         rec->prot     = fle->f.r.r_ip_p;
  673         rec->tos      = fle->f.r.r_tos;
  674         rec->dst_mask = fle->f.dst_mask;
  675         rec->src_mask = fle->f.src_mask;
  676 
  677         /* Not supported fields. */
  678         rec->src_as = rec->dst_as = 0;
  679 
  680         if (header->count == NETFLOW_V5_MAX_RECORDS)
  681                 return (1); /* end of datagram */
  682         else
  683                 return (0);     
  684 }
  685 
  686 /* Periodic flow expiry run. */
  687 void
  688 ng_netflow_expire(void *arg)
  689 {
  690         struct flow_entry       *fle, *fle1;
  691         struct flow_hash_entry  *hsh;
  692         priv_p                  priv = (priv_p )arg;
  693         item_p                  item = NULL;
  694         uint32_t                used;
  695         int                     i;
  696 
  697         /*
  698          * Going through all the cache.
  699          */
  700         for (hsh = priv->hash, i = 0; i < NBUCKETS; hsh++, i++) {
  701                 /*
  702                  * Skip entries, that are already being worked on.
  703                  */
  704                 if (mtx_trylock(&hsh->mtx) == 0)
  705                         continue;
  706 
  707                 used = atomic_load_acq_32(&priv->info.nfinfo_used);
  708                 TAILQ_FOREACH_SAFE(fle, &hsh->head, fle_hash, fle1) {
  709                         /*
  710                          * Interrupt thread wants this entry!
  711                          * Quick! Quick! Bail out!
  712                          */
  713                         if (hsh->mtx.mtx_lock & MTX_CONTESTED)
  714                                 break;
  715 
  716                         /*
  717                          * Don't expire aggressively while hash collision
  718                          * ratio is predicted small.
  719                          */
  720                         if (used <= (NBUCKETS*2) && !INACTIVE(fle))
  721                                 break;
  722 
  723                         if ((INACTIVE(fle) && (SMALL(fle) || (used > (NBUCKETS*2)))) ||
  724                             AGED(fle)) {
  725                                 TAILQ_REMOVE(&hsh->head, fle, fle_hash);
  726                                 expire_flow(priv, &item, fle);
  727                                 used--;
  728                                 atomic_add_32(&priv->info.nfinfo_inact_exp, 1);
  729                         }
  730                 }
  731                 mtx_unlock(&hsh->mtx);
  732         }
  733 
  734         if (item != NULL)
  735                 return_export_dgram(priv, item);
  736 
  737         /* Schedule next expire. */
  738         callout_reset(&priv->exp_callout, (1*hz), &ng_netflow_expire,
  739             (void *)priv);
  740 }

Cache object: 8dfaebe9fa3f7f9a49b58d27aefbc2df


[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]


This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.